Passivity And Low-Power Control Of Delayed Memristive Neural Networks

Memristor is an important circuit element to realize the function of artificial synapse,which makes the hardware realization of neural network possible.Therefore,it is significant to model the memristive neural network and analyze its dynamic behavior,which can provide theoretical support for the realization and construction of the neural network circuit system.Passivity and synchronization are both major dynamic behaviors in complex networks,the related basic behavior analysis and design of low-power controllers have been received much attention.The thesis mainly studies the passivity and passification of delayed memristive neural networks,as well as sliding-mode synchronization of delayed memristive neural networks under the low-power control strategy.Through considering the multi-proportional delays and impulse that appear in the actual circuit,this thesis constructs memristive recurrent neural networks with multiproportional delays and impulse,which is more practical.In this thesis,the conditions for realizing the passivity of the system are analyzed and a linear controller is proposed based on those conditions.Firstly,the problem of discontinuity at the right end of the system is solved by set-valued mapping and differential inclusion,and the memristor-based switching system is converted into a general network model.Secondly,this thesis constructs a new Lyapunov functional and combines inequality techniques to derive a new criterion for making the system passive,and effectively relaxes the condition that all matrices in the Lyapunov functional need to be positive definite,which reduces the conservatism of the results.Then,a linear external controller is designed based on the passive criterion,and the passification of the system is realized.Finally,the validity of the conclusions of this thesis has been fully verified by experimental results.Considering that energy consumption is an important evaluation index of system performance,in order to reduce power consumption,an event-triggered sliding mode controller based on continuous/periodic sampling algorithm was designed in this thesis to realize the synchronization of delayed memristive neural network.First,this thesis presents a sliding-mode controller triggered by continuous sampling events to synchronize the system.Secondly,by analyzing the behavior of the system under continuous sampling,this thesis has obtained the stable motion region of the system and the lower bound of the time interval between two consecutive updates of the controller,which effectively avoids the Zeno phenomenon.Finally,based on the given interval lower bound,this thesis proposes a periodic sampling algorithm with lower power consumption.By analyzing and discussing the simulation results,both the continuous sampling and the periodic sampling algorithm can synchronize the system,but the periodic sampling algorithm has fewer updates of the system controller,which effectively reduces the computational burden of the system and has more practical application value.The research content of this thesis enriches the research results of behavior analysis of memristive neural network,such as passivity and sliding mode synchronization,and provides more theoretical support for the research and application of memristive neural network in artificial intelligence.